Martin (1992: 140):
Here, with reference structures, it is the dependency approach that will be pursued. The reason for this is that reference structures are essentially covariate ones. Where endophoric, phoric items depend semantically on the items they presume, but need not be grammatically related to them. And presuming items can themselves be presumed. The covariate structures deriving from phoric options in the IDENTIFICATION system will be referred to as reference chains (closely related to Hasan's identity chains; see Hasan 1984, 1985).
 No argument has been made as to why these identity relations between instantial participants are textual structures. They are simply asserted to be so.
 Dependency relations are features of the logical metafunction that obtain between rank units in complexes (forms). Here they are being used for the textual metafunction, rather than the logical, which is a major theoretical inconsistency. The fact that they are not applied to rank units in complexes is a further (internal) theoretical inconsistency.
 To be clear, the relation between structure and system is realisation (elaboration + identity).